Running experiments with tsexperiment
Now it's time to run some workloads. As of TSLoad 0.2 it's the only way to run experiment because it doesn't support multi-agent mode, so first of all you will need to create a directory containing experiment.json file with experiment's configuration which defines threadpools and workloads. We omit this step in this guide because it was discussed earlier.
If you do so - let's check that config is parseable by TSLoad. I will use sample experiment shipped with TSLoad:
$ ./bin/tsexperiment -e var/tsload/sample/ show name : test root : var/tsload/sample/ directory : runid : -1 status : NOTCFG THREADPOOL NTHRS QUANTUM DISP test_tp 2 1s random WORKLOAD TYPE THREADPOOL RQSCHED test1 busy_wait test_tp think test2 busy_wait test_tp iat test3 busy_wait - - chained: test2
As you can see there are one threadpool called test_tp and three workloads attached to it.
$ ./bin/tsexperiment -e var/tsload/sample/ run Created const steps generator for workload 'test1' with N=20 R=2000 Loaded steps file 'var/tsload/sample/test2.tss' for workload 'test2' === CONFIGURING EXPERIMENT 'test' RUN #3 === Configured threadpool 'test_tp' with 2 threads and 1s quantum Workload 'test1' configuring (0): Workload 'test1' successfully configured (100): Workload 'test2' configuring (0): Workload 'test2' successfully configured (100): Workload 'test3' configuring (0): Workload 'test3' successfully configured (100): === STARTING WORKLOADS @01:43:22 14.02.2014 === Starting workload 'test1': provided 15 steps Starting workload 'test2': provided 15 steps Starting workload 'test3': provided 0 steps ... (some output is ommitted) ... === UNCONFIGURING EXPERIMENT 'test' RUN #3 === ... (some output is ommitted) ...
Experiment consists of several runs which could be listed using list subcommand:
$ ./bin/tsexperiment -e var/tsload/sample/ list ID NAME STATUS DATE HOSTNAME ... 3 test FINISHED 01:43:22 14.02.2014 panther.mylocal
Timestamp in column DATE is a scheduled start time of experiment's workloads, so if you use some external monitoring tool, use it as baseline to correlate values.
Let's take closer look to subdirectory created by tsexperiment for test run number 3:
$ ls -l var/tsload/sample/test-3/ total 1662 -rw-r--r-- 1 myaut users 1715 Feb 13 21:41 experiment.json -rw-r--r-- 1 myaut users 1411 Feb 13 21:40 experiment.json.old -rw-r--r-- 1 myaut users 870 Feb 13 21:40 test1-schema.json -rw-r----- 1 myaut users 2564096 Feb 13 21:41 test1.tsf -rw-r--r-- 1 myaut users 870 Feb 13 21:40 test2-schema.json -rw-r----- 1 myaut users 22528 Feb 13 21:41 test2.tsf -rw-r--r-- 1 myaut users 51 Feb 13 21:41 test2.tss -rw-r--r-- 1 myaut users 870 Feb 13 21:40 test3-schema.json -rw-r----- 1 myaut users 9728 Feb 13 21:41 test3.tsf -rw-r--r-- 1 myaut users 3402 Feb 13 21:41 tsexperiment.out
As you can see there is copy of experiment.json and test2.tss configuration files, file called tsexperiment.out that contains output of tsexperiment utility and trace files with tsf exception (which means TS File). Because it is binary-formatted there is a schema file nearby that helps TSLoad to parse that file.
You may alter some of experiment parameters on the fly. For example let's increase number of requests per step for workload 'test1' to 8000 and 12000. Let's dump experiment config in name-value format:
$ ./bin/tsexperiment -e var/tsload/sample/ show -l name=test steps:test1:num_steps=20 steps:test1:num_requests=2000 ... (some output is ommitted)
Parameter called 'steps:test1:num_requests' is what we need. It would be also useful to specify -n option to create custom run name. Now rerun experiment with these values:
$ ./bin/tsexperiment -e var/tsload/sample/ run -s steps:test1:num_requests=8000 -n run-8k ... (some output is ommitted)
If you run command like top or mpstat you will see that processor consumption for tsexperiment is increased. You will also may want to reproduce experiment precisely - in this case you need to run tsexperiment with -T option:
$ ./bin/tsexperiment -e var/tsload/sample/ run -T 3 -n trace ... (some output is ommitted)
So when we will rerun 'list' subcommand, we will se new experiment runs:
$ ./bin/tsexperiment -e var/tsload/sample/ list ID NAME STATUS DATE HOSTNAME ... 11 run-8k FINISHED 02:37:11 14.02.2014 panther.mylocal 12 run-12k FINISHED 02:37:59 14.02.2014 panther.mylocal 13 test FINISHED 02:39:15 14.02.2014 panther.mylocal
Write these numbers down because we will need them onto next step.